On Thu, Apr 06, 2017 at 03:10:41PM +0200, Christian Affolter wrote: > Hi Brian, > > thanks a lot for taking the time to help. > > On 06.04.2017 14:51, Brian Foster wrote: > > On Wed, Apr 05, 2017 at 03:08:03PM +0200, Christian Affolter wrote: > > > Hi everyone, > > > > > > I'm having a system with a 15 TiB XFS volume with user and project quotas, > > > which recently crashed. > > > > > > After a successful xfs_repair run, I'm trying to mount the file system > > > again. According to dmesg a quotacheck is needed (XFS (vdc): Quotacheck > > > needed: Please wait.), for about 20 minutes I see expected read-activity on > > > the device within iotop, then it suddenly stops but the mount command still > > > hangs in uninterruptable sleep state (D+) while the system remains > > > completely idle (I/O and CPU wise). > > > > > > [...] > > > > > Apr 5 10:33:48 sysresccd kernel: XFS (vdc): Mounting V5 Filesystem > > > Apr 5 10:33:48 sysresccd kernel: XFS (vdc): Ending clean mount > > > Apr 5 10:33:53 sysresccd kernel: XFS (vdc): Unmounting Filesystem > > > Apr 5 10:34:00 sysresccd kernel: Adding 4193276k swap on /dev/vda2. Priority:-1 extents:1 across:4193276k FS > > > Apr 5 10:34:35 sysresccd kernel: XFS (vdc): Mounting V5 Filesystem > > > Apr 5 10:34:35 sysresccd kernel: XFS (vdc): Ending clean mount > > > Apr 5 10:34:35 sysresccd kernel: XFS (vdc): Quotacheck needed: Please wait. > > > Apr 5 10:34:43 sysresccd su[2622]: Successful su for root by root > > > Apr 5 10:34:43 sysresccd su[2622]: + /dev/tty2 root:root > > > Apr 5 12:03:49 sysresccd kernel: sysrq: SysRq : Show Blocked State > > > Apr 5 12:03:49 sysresccd kernel: task PC stack pid father > > > Apr 5 12:03:49 sysresccd kernel: mount D ffff880310653958 0 2611 2571 0x20020000 > > > Apr 5 12:03:49 sysresccd kernel: ffff880310653958 ffff880310653958 0000000500370b74 ffff880002789d40 > > > Apr 5 12:03:49 sysresccd kernel: ffff880310654000 ffff8803111da198 0000000000000002 ffffffff817fbb2a > > > Apr 5 12:03:49 sysresccd kernel: ffff880002789d40 ffff880310653970 ffffffff817f9f58 7fffffffffffffff > > > Apr 5 12:03:49 sysresccd kernel: Call Trace: > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff817fbb2a>] ? usleep_range+0x3a/0x3a > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff817f9f58>] schedule+0x70/0x7e > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff817fbb59>] schedule_timeout+0x2f/0x107 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff81170dcd>] ? __kmalloc+0xeb/0x114 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff81389780>] ? kmem_alloc+0x33/0x96 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff810b58f0>] ? arch_local_irq_save+0x15/0x1b > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff817fc3f7>] ? _raw_spin_unlock_irqrestore+0xf/0x11 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff817fa72d>] do_wait_for_common+0xe4/0x11a > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff817fa72d>] ? do_wait_for_common+0xe4/0x11a > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff810a315c>] ? wake_up_q+0x42/0x42 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff817fa7d9>] wait_for_common+0x36/0x4f > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff817fa80a>] wait_for_completion+0x18/0x1a > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff8139a223>] xfs_qm_flush_one+0x42/0x7f > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff8139a586>] xfs_qm_dquot_walk.isra.8+0xc1/0x106 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff8139a1e1>] ? xfs_qm_dqattach_one+0xe3/0xe3 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff8139bd27>] xfs_qm_quotacheck+0x131/0x252 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff8139bedf>] xfs_qm_mount_quotas+0x97/0x143 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff813840a1>] xfs_mountfs+0x587/0x6b1 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff813868ea>] xfs_fs_fill_super+0x411/0x4b5 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff811846d1>] mount_bdev+0x141/0x195 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff813864d9>] ? xfs_parseargs+0x8e8/0x8e8 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff81168bff>] ? alloc_pages_current+0x96/0x9f > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff8138502b>] xfs_fs_mount+0x10/0x12 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff811851f3>] mount_fs+0x62/0x12b > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff81199646>] vfs_kern_mount+0x64/0xd0 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff8119bccd>] do_mount+0x8d5/0x9e6 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff811c07b1>] compat_SyS_mount+0x179/0x1a5 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff810039bb>] do_syscall_32_irqs_off+0x52/0x61 > > > Apr 5 12:03:49 sysresccd kernel: [<ffffffff817feee6>] entry_INT80_compat+0x36/0x50 > > > > This looks like a known quotacheck vs. reclaim deadlock. Quotacheck is > > in the phase where it has completed all of the updates of in-core > > accounting and now needs to flush the in-core xfs_dquot structures to > > their associated backing buffers and submit them for I/O. The flush > > sequence is stuck waiting on an xfs_dquot flush lock, however. > > > > The problem is that quotacheck holds the underlying buffers and then > > does a blocking xfs_dqflock(), which is effectively an inverse locking > > order from xfs_dquot memory reclaim. If reclaim runs during the previous > > quotacheck phase, it acquires the flush lock and fails to submit the > > underlying buffer for I/O (because quotacheck effectively holds/pins the > > buffer). When the quotacheck flush gets to the dquot, it thus waits > > indefinitely on a flush lock that will never unlock. > > > > If you're comfortable building a kernel from source, you could try the > > following patch and see if it helps quotacheck to complete: > > > > http://www.spinics.net/lists/linux-xfs/msg04485.html > > > > (Note that I've not been able to get this patch merged.) > > Shall I only apply "[PATCH 2/3] xfs: push buffer of flush locked dquot to > avoid quotacheck deadlock" or are all patches of this series required? > > Which kernel version would you recommend for applying the patch? > You could apply patches 1-2 of that series. Patch 3 was a hack/experiment. Definitely do not apply that one. I think they should apply fine to any v4.10 or newer kernel. I'm not sure that quotacheck changes all that often so you could probably pull them into an older kernel as well, but you'd have to try to apply them to know for sure. Brian > > > Otherwise, you'd probably have to go a ways back to a kernel before this > > regression was introduced. That looks like v3.4 according to the code > > (but it's also possible other changes may have prevented this state > > between then and with more recent kernels, particularly since it seems > > to have only turned up more recently). > > > > Brian > > Thanks again, > Chris > -- To unsubscribe from this list: send the line "unsubscribe linux-xfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
